AUV Pipeline Following using Reinforcement Learning

Konferenz: ISR/ROBOTIK 2010 - ISR 2010 (41st International Symposium on Robotics) and ROBOTIK 2010 (6th German Conference on Robotics)
07.06.2010 - 09.06.2010 in Munich, Germany

Tagungsband: ISR/ROBOTIK 2010

Seiten: 8Sprache: EnglischTyp: PDF

Persönliche VDE-Mitglieder erhalten auf diesen Artikel 10% Rabatt

Fjerdingen, Sigurd A.; Kyrkjebø, Erik; Transeth, Aksel A. (SINTEF ICT, 7465 Trondheim, Norway)

This paper analyzes the application of several reinforcement learning techniques for continuous state and action spaces to pipeline following for an autonomous underwater vehicle (AUV). Continuous space SARSA is compared to the actor-critic CACLA algorithm, and is also extended into a supervised reinforcement learning architecture. A novel exploration method using the skew-normal stochastic distribution is proposed, and evidence towards advantages in the case of tabula rasa exploration is presented. Results are validated on a realistic simulator of the AUV, and confirm the applicability of reinforcement learning to optimize pipeline following behavior.